Face Recognition Research Based On Transform Domain And Local Direction Pattern

With the development of biometrics,face recognition technology has been widely used,especially in public safety and production.Compared with other biometric technologies,face recognition is convenient,efficient,and safe.It lays a foundation for its successful application and popularization,and it has become a hot topic of current research.However,face recognition is still affected by factors such as illumination,expression,posture and occlusion in the actual application scene.This makes face recognition a huge challenge and needs further research.The multi-scale geometric analysis method reveals the concrete expression of high-dimensional functions in the context of optimal approximation.This method is a simulation of the human eye's perception process of things,reveals different feature sub-bands,has rich resolution,and can better handle local and overall relationships.Therefore,when the technology is applied to the image processing field,it can be obtained.Better fusion effect.Local Directional Pattern(LDP)is a kind of excellent local feature extraction method.Because its feature extraction method is simple,it is not enough to extract the total face information.The face recognition algorithm for local direction mode is studied.An improved face recognition algorithm is proposed.The main research contents of the thesis include:(1)Aiming at the problem that the recognition rate caused by the change of posture,illumination and expression in face recognition is not high,a Non-subsampled Contourlet transform(NSCT)and an absolute value center symmetric local direction(Absolute Center of Symmetry Local Directional Patterns,ACSLDP)combined face recognition method.Firstly,the NSCT processing is performed on the face image after simple pre-processing,and the sub-band images with different directions in different directions are obtained.Then the sub-band image is calculated to improve the feature extraction of the local direction mode algorithm to obtain the ACSLDP feature,and then for each frame.The image is divided into blocks,each sub-image is weighted by information entropy,the histogram feature information is statistically analyzed,and all sub-image histograms are connected as the facial features of the whole algorithm,and are identified by the nearest neighbor classifier.Tested on ORL,YALE and CAS-PEAL-R1 face database,the experimental results show that the face image has strong recognition ability,the feature dimension is small,and it is robust to posture,illumination and expression changes.(2)In order to make the face image information more fully extracted,face recognition combined with Non-subsampled Shearlet transform(NSST)and Absolute Double Center Symmetry Local Pattern(ADCSLP)is proposed.The lower 4 bits of the ADCSLP code arecomposed of a central symmetric local binary mode coding rule,and the upper 4 bits of the ADCSLP code are composed of absolute value type central symmetric local direction mode coding rules.Compared with LDP,CSLDP,DLDP and GCSLDP algorithms,the ADCSLP algorithm considers both the original data space information and the gradient space information,making full use of the face feature information of the original space and the gradient space.On the YALE and AR face databases,the face algorithm combining NSST and ADCSLP has achieved high recognition results.